1 Cherenkov Telescope Array: a production system prototype L. Arrabito 1 C. Barbier 2, J. Bregeon 1, A. Haupt 3, N. Neyroud 2 for the CTA Consortium 1 LUPM CNRS-IN2P3 France 2 LAPP CNRS-IN2P3 France 3 DESY EGI Community Forum, November 2015, Bari

Outlook: CTA in a nutshell CTA computing model: o Data volume o Data flow o Data processing CTA production system prototype: o Current MC simulations and Analysis o DIRAC for CTA o Resource usage: Future developments Conclusions 2

CTA in a nutshell 3 Scientific goals: Cosmic rays origins High Energy astrophysical phenomena Fundamental physics and cosmology CTA (Cherenkov Telescope Array) is the next generation instrument in VHE gamma-ray astronomy 2 arrays of Cherenkov telescopes (North and South hemispheres) 10x sensitivity with respect to current experiments Consortium of ~1200 scientists in 32 countries Operate as an observatory Site locations decided for further negotiations this year: o North site: La Palma, Spain o South site: Paranal ESO, Chile Currently in ‘Pre-construction’ phase ( ) Operations will last ~30 years

CTA Computing: data volume Raw-data rate CTA South: 5.4 GB/s CTA North: 3.2 GB/s 1314 hours of observation per year Raw-data volume ~40 PB/year ~4 PB/year after reduction Total volume ~27 PB/year including calibrations, reduced data and all copies 4

CTA Computing: data flow 5

CTA Computing: data processing From raw-data to high level science data: Several complex pipelines (Calibration, Reconstruction, Analysis, MC, etc.) Assume 1 full re-processing per year Reconstruction Pipeline shower reconstruction step 6

CTA is now in ‘Pre-construction’ phase Massive MC simulations and Analysis running for 3 years o ‘Prod2’ ( )  Characterize all site candidates to host CTA telescopes to determine the one giving the best instrument response functions  4.6 billion events generated for each site candidate, 2 different zenith angles and 3 telescope configurations  8 full MC campaigns (5 sites for the South and 3 for the North) o ‘Prod3’ (2015) in progress:  For the 2 selected sites: study the different possible layouts of telescope arrays, pointing configurations, hardware configurations, etc.  800 telescope positions, 7 telescope types, multiple possible layouts, 5 different scaling  Run 3 different Analysis chains on the simulated data  Each one processing about 500 TB and 1 M of files for 36 different configurations Computing is already a challenge! MC simulations and Analysis 7

Resources: Dedicated storage: o Disk: 1.3 PB in 6 sites: CC-IN2P3, CNAF, CYFRONET, DESY, GRIF, LAPP o Tape: 400 TB in 3 sites CPU: 8000 cores available on average CTA Virtual Organization: Active since EGI sites in 7 countries and 1 ARC site About 100 members DIRAC for CTA: Dedicated DIRAC instance composed of 4 main servers at CC-IN2P3 and PIC Several DIRAC Systems in use CTA-DIRAC software extension MC simulations and Analysis 8 Use of an existing and reliable e-Infrastructure: EGI grid Use of DIRAC for Workload and Data Management

Computing Model: Use of CTA-DIRAC instance to access grid resources MC simulation uses CPU resources of all 20 grid sites supporting the CTA VO Output Data are stored at 6 main SE MC Analysis takes place at the 6 sites where MC data are stored Computing Operations (small team of people): Receive production requests from MC WP (nb of events to be generated, sw to install, etc.) Adjust the requests according to the available resources Negotiate resources with grid sites on a yearly basis Run the productions and perform data management operations (removal of obsolete datasets, data migration, etc.) Support users to run their private MC productions and analysis DIRAC servers administration Development of CTA-DIRAC extension MC simulations and Analysis 9

DIRAC for CTA: main Systems in use Workload Management System o Job brokering and submission (pilot mechanism) o Integration of hetereogenous resources (CREAM, ARC) o Central management of CTA VO policies Data Management System o All data operations (download, upload, replication, removal) o Use of the DIRAC File Catalog (DFC) as Replica and Metadata catalog Transformation System o Used by production team to handle ‘repetitive’ work (many identical tasks with a varying parameter), i.e. MC productions, MC Analysis, data management operations (bulk removal, replication, etc.) 10

11 Metadata selection Catalog browsing Query result In use since 2012 in parallel with LFC. Full migration to DFC in summer 2015 More than 21 M of replicas registered About 10 meta-data defined to characterize MC datasets DIRAC for CTA: DIRAC File Catalog Query example: cta-prod3-query --site=Paranal -- particle=gamma --tel_sim_prog=simtel --array_layout=hex --phiP=180 --thetaP=20 --outputType=Data Typical queries return several hundreds of thousands of files DFC web interface

DIRAC for CTA: Transformation System Workload Management System Production Manager MC jobs Transformation System DFC Plugins Transformations Transformation Agent Workflow Task Agent Database table Agent Tasks InputData Agent Files InputDataQuery Request Management System Request Task Agent Transformation System Architecture The Production Manager defines the transformations with meta-data conditions and ‘plugins’ InputData Agent queries the DFC to obtain files to be ‘transformed’ Plugins group files into tasks according to desired criteria Tasks are created and submitted to the Workload or Request Management System 12 Transformation Monitoring

Resource usage: Use of about 20 grid sites concurrent jobs for several weeks Users analysis also running in parallel (private simulations, analysis) More than 7.7 M jobs executed Start prod2 SAC Tenerife Users Analysis Paranal El Leoncito Aar US SPM Start prod3 13 MC Paranal MC Analysis Running jobs by site Prod M HS06 hours 640 TB Prod M HS06 hours 785 TB 5000 jobs

Resource usage: About 2.6 PB processed (MC Analysis) Throughput of MB/s during ‘prod3’ PB Processed data Throughput 1 GB/s

Until now, we have been using almost all DIRAC functionalities For long term operations CTA is developing several systems: o Archive system to store and manage CTA data o Pipeline framework to handle all CTA applications o Data model to define the whole meta-data structure of CTA data o Science gateway for end-users data access o DIRAC will be used for the Workload and Production Management Future developments: o Develop interfaces between DIRAC and the other CTA systems o Improvement of the DIRAC Transformation System toward a fully data- driven system (next slide) o Improve CTA-DIRAC hardware setup Future developments 15

Toward a fully data-driven Transformation System Developments in progress: When new files are registered, a filter based on meta-data is applied to send them on the fly to their matching transformation No need anymore to perform heavy queries of the Archive 16 Workload Management System Production Manager Pipeline Transformation System Archive Catalog Plugins Transformations Transformation Agent Workflow Task Agent Database table Agent Tasks InputData Agent Files InputDataQuery Files with meta-data Filter by meta- data

17 Improve CTA-DIRAC hardware setup DIRAC is based on a Service Oriented Architecture Each DIRAC System is composed of Services, Agents and DBs CTA-DIRAC instance is a rather modest installation, composed of: 3 core servers: o 1 server running all Services (except DM) and a few Agents (4 cores) o 1 server running all Agents except TS (2 cores) o 1 server running the DataManagement System and Transformation Agents (16 cores) 2 MySQL servers: o 1 server hosting all DBs except FileCatalogDB o 1 server hosting the FileCatalogDB 1 web server Observed high load on the core servers when running several ‘productions’ in parallel Need to add more servers, but also optimize the component distribution on the servers

Conclusions 18 CTA will produce about 27 PB/year and it will operate for ~30 years A production system prototype based on DIRAC has been developed for MC simulation productions: o Required minimal development of a CTA-DIRAC extension o Minimal setup and maintenance of the instance running all the components, mostly the work of 1 person with help of DIRAC team and sys admins upon request Successfully used during last 3 years for massive MC simulations and analysis: o > 1.6 PB produced o ~ 2.6 PB processed o > 20 M files registered in the Catalog Future developments: o Improve production system for real data processing pipeline  Build interfaces between DIRAC and other CTA systems (Archive, Pipeline, etc.)  Further develop DIRAC Transformation System toward a fully data- driven system